Bug#1050220: mutter: autopkgtest regression on s390x: tests time out after 5 minutes

[Simon McVittie]

Source: mutter
Severity: normal
Version: 44.0-2
Tags: help
X-Debbugs-CC: debian-s390 at lists.debian.org
User: debian-s390 at lists.debian.org
Usertags: s390x

Recent versions of the mutter compositor/window manager library fail their
"as-installed" tests on s390x.

Steps to reproduce:

* have a Debian testing/unstable environment
* `apt install gnome-desktop-testing mutter-${v}-tests`
  (where ${v} is 11, 12 or 13 for mutter 43.x, 44.x or 45.x respectively)
* `ginsttest-runner mutter-`
  (you can add other options from its --help if they are useful to you)

Or run the "installed-tests" autopkgtest from the mutter source package,
which automates this.

Expected result:

* 20+ test-cases are run (the exact number varies between versions)
* A summary at the end shows something like this:
  SUMMARY: total=31; passed=31; skipped=0; failed=0; user=16.6s; system=3.7s; maxrss=150496
  (possibly with "# " prepended, if you are using ginsttest-runner --tap)

Actual result:

With 43.x, the expected result.

With 44.x, many tests run for 5 minutes (300 seconds) each, with no
useful output, and then are killed by ginsttest-runner with signal 9:

    4949s # Executing: mutter-12/closed-transient-no-input-no-take-focus-parent.test
    4949s not ok - Test timed out after 300 seconds
    4949s # FAIL: mutter-12/closed-transient-no-input-no-take-focus-parent.test (Child process killed by signal 9)
    4949s not ok - mutter-12/closed-transient-no-input-no-take-focus-parent.test
    4949s # SUMMARY: total=35; passed=19; skipped=0; failed=16; user=8.6s; system=1.8s; maxrss=146164

Since version 44.3-7, this test failure is ignored on s390x, on the basis
that getting GNOME 44 into unstable for users of architectures where
desktop/laptop systems exist seems like it should be a higher priority
for the Debian project than making a 3D-accelerated compositing manager
available on mainframe computers that do not typically have any graphical
output at all. The regression is presumably still a bug, though.

The build-time tests are currently also skipped on s390x.

None of the GNOME maintainers own a $15K+ mainframe or have any particular
interest in running graphical software on one, so we are not intending
to work on this issue further. If this is to be investigated, someone
who values the ability to run the GNOME desktop on s390x should take
responsibility.

s390x has been a repeated time-sink for GNOME maintenance, and I am not
aware of any reasons why mutter/gnome-shell would ever be practically
useful on a mainframe, so I am very tempted to make mutter artificially
unbuildable on that architecture and ask for older mutter, gnome-shell
and related packages to be removed from the archive. This would not
affect the ability to run individual graphical programs from GNOME on
s390x via X11 forwarding (if anyone does that), but it would prevent
the full GNOME desktop from being installed or run on s390x.

Thanks,
    smcv